While I agree that 2 hour samples are less accurate, I doubt that 3-4 past weeks data is somewhat useful for real-time performance troubleshooting, this data is definitely useful for trending/capacity planning etc.
For example, If you have a performance issue and it is causing you some real trouble, the first thing you would do is check up past hour performance, then you'll compare previous day, then you'll shift your focus to a previous week. Going further than past two weeks in order to isolate the problem doesn't sound right to me, because this would mean that this issue stayed unnoticed/hidden for two weeks. And in this case our initial performance issue doesn't look like a real issue because you've lived with it for more than two weeks.
Does it make any sense?